T Cell–Inspired Algorithms - Immunological Computation: Theory and Applications - page 62

h e purpose of NS is to test for tolerance of self-cells. T cells that recognize the
combination of MHC and self-peptides fail this test. h is process can be viewed
as a fi ltering of a big diversity of T cells; only those T cells that do not recognize
self-peptides are kept (Kappler et al., 1987).

4.2 Negative Selection Algorithms

Forrest et al. (1994) proposed a computational model of self/nonself discrimination,
which is called the “NSA or NS algorithm.” h is algorithm models the T cell
maturation process that occurs in the thymus. Several variations of NSAs have been
proposed after the original version was introduced (Forrest et al., 1994); however, the
main features of the original algorithm still remain. Particularly, the goal of NS is to
cover the nonself space with an appropriate set of detectors (shown in Figure 4.1).
Two important aspects of an NSA are as follows:
1. h e target concept of the algorithm is the complement of a self-set.
2. h e goal is to discriminate between self and nonself patterns, while only self-
samples are available (one-class learning; Tax, 2001).

h ere are two steps in NSAs as follows: “detector generation” and “nonself
detection.” In the fi rst step, a set of detectors is generated by some random-
ized process that uses a collection of self as the input. Candidate detectors that
match any of the self-samples are eliminated, whereas unmatched ones are kept.

Self

Self

Figure 4.1 Illustration of the self and nonself regions.